20:00Machine-translated from Azerbaijani by AIRead in Azerbaijani
Meta Unveils Muse Wonder AI Models With Long-Term Memory
Meta has unveiled a new family of artificial intelligence models called Muse Wonder , designed to perform complex tasks in everyday life and professional settings. The company views the development as an important step toward personal superintelligence. The goal is to create AI that understands the user’s habits and goals, remembers important information, and can work independently to complete a task.
The new family includes two models: Muse Wonder and the more powerful Muse Wonder Full. The base model is already available through the Meta Model API and the Muse Code tool. The more powerful version is currently available to Muse Code subscribers. In the future, the Muse Wonder model will also be integrated into the Muse service. There, it will be used to create various types of content and perform tasks requiring more advanced reasoning.
One of the main innovations in the Muse Wonder models is an active memory system called Active Memory. This technology allows the AI to remember important information about the user and reuse it when needed. As a result, the user may not have to explain the same information repeatedly. By taking previous context, personal preferences, and stated goals into account, the AI can produce more relevant responses and continue working on tasks.
Another important technology is called Active Focus. It helps the model work on complex tasks for extended periods, try different solutions, overcome obstacles, and adapt to new information. If additional information is needed while completing a task, the model can ask the user for clarification. It can also coordinate with the user when making important decisions and taking important actions.
According to Meta, the Muse Wonder models’ programming skills and ability to work with professional information have also been improved. They can perform tasks intended for fields that require specialized knowledge, such as law, finance, and medicine.
The models were trained on long-term tasks requiring them to work consistently, verify results, and correct errors. According to company employees, the new models achieve strong results in relevant tests and perform better at solving tasks they have not previously encountered. This approach could turn Muse Wonder into a tool with broader capabilities than a chatbot that simply answers questions. The model’s goal is not only to provide instructions but also to continue the work process until a result is achieved.
Because personal AI can gain access to the user’s messages, calendar, and documents, protecting data is particularly important. Meta says the Muse Wonder models were designed to protect confidential information and resist malicious instructions.
The company also emphasizes that additional security measures have been strengthened to reduce the risk of the models being used for dangerous purposes.